7

As the title says, I have two power supplies in a server on the same circuit. The circuit is connected to surge protection, power filtering, and UPS. The only reason I use both power supplies is because:

  1. I'd like to be prepared against a failing power supply, and
  2. I don't want to see alarms (power supply red).

Is there anything wrong with this?

msrd0
  • 240
  • 3
  • 13
  • 5
    In fact I'd connect only one of the power supplies to UPS, and the other only to surge protection - this eliminates failed UPS without reducing the effectiveness of the UPS in case of power failure. – Hagen von Eitzen Sep 23 '16 at 16:34

3 Answers3

21

No, not really. This is a perfectly good way to protect against a power supply failing, as long as it doesn't fail in such a way as to draw massive amounts of current (very unusual).

Normally people will put them on different circuits to also protect against blown breakers, and use different power sources for those circuits sometimes to protect against outages and blown UPSes. If you are only worried about a failing PSU you don't need to worry about that.

Falcon Momot
  • 25,244
  • 15
  • 63
  • 92
  • Do you have a source about connecting different power sources (assuming no common base source), maybe from some physics site? As a non expert it's something i would have avoided, so it's interesting what could happen and why it does not happen. – allo Sep 25 '16 at 22:29
  • No, I don't. This is pretty standard practice though, very far removed from theoretical physics. Nearly every server I have seen has each PSU connected to a different PDU, often at least on a separate circuit. With a 3-phase supply, one might connect server 1 to phase 1 and 2, server 2 to 2 and 3, and server 3 to 1 and 3, so that with one phase working at least 2 are up, etc. – Falcon Momot Sep 26 '16 at 00:49
  • 1
    @allo Probably all the sources use earth (ground) as their 0V reference. – derobert Sep 26 '16 at 17:52
1

A lot depends on how much power is available on the circuit, and how much your servers need. If you only have 1 phase (small computer room), you have no choice. As you get close to the rating of the circuit, expect more power supply failures and circuit breaker pops. Try to not to exceed 80% of maximum circuit amp rating. Consult several licensed, commercial electricians to design your setup if you want to do a pro job.

In a small office, your panel may have residential split phase power - 120V on most breakers, but some 208V breakers, too. This is sort of like having 2 phases - 1 phase is 180 degrees out from the other. The phase difference doesn't hurt your gear one bit.

In a bigger datacenter, you often have a PDU or power panel that has 3 phases, 120 degrees from peak to peak. When we talk of balancing phases, we mean balancing the current on each phase, not the phase angle.

Balance the loads by putting different power supplies on different phases. This avoids filling up one phase with loads too quickly. When circuits are loaded near max, balancing also helps prevent accidental breaker trips when multiple devices power up due to inrush current. It also prevents server outages if 1 phase's breaker does indeed trip due to a wiring fault or catastrophic power supply failure. Do the arithmetic - in a redundant setup, you must run each phase less than 40% of load to allow 1 phase to die, and still have some room for inrush.

If you have multiple phases, label your breakers and power outlets under the floor and your power strips in your cabinets with the phases. If you don't have metered power panels or UPSes, count the rated amps of each supply, but bear in mind that server vendors offer options on whether 1 supply is idle or if both supplies share the load. My opinion is, share the load. If you want to geek it to the max, you can use actual DC loads on each supply to calculate. This is why people buy metered UPSes and PDUs - so much easier to eyeball.

As far as totally different sources of power, the only place I've seen that is in a hospital after a hurricane Katrina, where power feeds in the neighborhood were unstable. They got separate feeds from separate commercial power substations, and used switchgear (big automatic switches - hundreds of amps @ 240 or 480V) to detect and switch over. Very expensive. Power did burp when we lost a source, but our UPSes smoothed it out. Lights would dim or go out completely, but the servers stayed up while PCs on non protected power rebooted.

It's important in such cases to have isolation transformers and well installed star grounds (all rack power grounds returning to a single ground point under the floor). Power switchover can send differential spikes down any phase and neutral, and the isolation transformers prevent the UPSes and servers from seeing most of such spike. You can still get sags and surges, though - longer rises or dips in power. Again, the UPSes protect you.

Low cost surge protectors die to protect the circuit - do not rely on them or be prepared to change them out.

https://en.wikipedia.org/wiki/Surge_protector

Power gear vendors, like APC, server and network gear vendors have white papers that explain more of the technical details, but you need to wade through a lot of marketing hype. They often don't help smaller sites much.

Here's a paper I googled up on problems caused by unbalanced phases:

http://datacentertrends.blogspot.com/2011/11/power-in-data-center-and-balancing-3.html

Foon
  • 31
  • 2
-2

Nope, nothing wrong and yes I'd check if the sources of your power supplies are far out of phase or not before connecting both. You can trust your good 'ol multimeter for this.

Most of our racks have deviations, some equipment (most switches and servers) can handle this better than others.

Walter
  • 1
  • 2
    Any source on limitations to phase difference? – dbanet Sep 23 '16 at 18:10
  • 2
    Phase difference should affect nothing. Ground will be the same, and the 60 Hz waveform will be rectified and filtered out by the time it's at the business end of the PSU. DC has no concept of frequency, and thus no concept of phase. All "hot" circuits are isolated from one another until the DC output. – Spooler Sep 25 '16 at 07:40